An Improved Level-3 Oceanic Rainfall Retrieval Algorithm for TRMM

The level-3 rainfall algorithm currently being used for the routine processing

of TRMM Microwave Imager (TMI) data into monthly rainfall totals is based on

accumulating histograms of a linear combination of the 19.35 and 21.3 GHz channels.

The rainfall totals are computed by assuming a lognormal distribution of rainrates

and adjusting the parameters of the lognormal distribution to predict the observed

histogram. This procedure fills in the contribution of rainrates too high or too

low to be measured by the TMI. The procedure is also configured to solve for the

brightness temperature in the limit of no rain thereby canceling some calibration

and modeling uncertainties.

The combination of the TRMM Precipitation Radar (PR) with the TMI has permitted

some refinements of the physics assumptions of the retrieval model. In particular,

the comparison of freezing levels inferred from the brightband in the PR data and

retrieved from the TMI uncovered the need for updating the water vapor spectroscopic

assumptions in the model.

A number of advantages would accrue from accumulating histograms of rain rate rather

than brightness temperature. The most notable of these advantages is the ability to

use all of the frequencies of the TMI to expand the dynamic range of the measurements

and reduce the need to fill in high and low rain rates via the lognormal assumption.

Previous attempts as tested in various algorithm intercomparisons have not worked

well. However, the experience developed in these intercomparisons has enabled the design

of an algorithm that avoids the problems that have been uncovered.

The new algorithm collects histograms of rain rate as derived from several different

channels, each valid over a limited dynamic range. The rain rates are smoothed to the

resolution of the 10.7 GHz channel so that the same set of lognormal parameters should

describe each of them. After correcting for offsets, a single histogram is constructed

for each 5 degree square using the most appropriate rain rate retrieval (i.e. frequencies)

for each pixel.

This new algorithm should, in principle, provide more accurate rainfall totals. The

available rain truth over the oceans has proved grossly inadequate for validating

rainfall retrievals and model based error analyses are necessary to establish the

uncertainties of the rainfall estimates.